Presentation

 < Day Day Up > 



The acquisition of knowledge derived from a set of data requires a presentation of the data's underlying structure to the user. The most common method is visual presentation, which is the focus of this section. However there is current research into the auditory presentation of data and the benefits of combining auditory with visual presentations (Barass, 1995). There are two classes of visual presentation: textual and graphical. Textual presentation methods are simpler but more constrained. Graphical techniques are more difficult to implement but show more promise as they facilitate discovery through incorporating human perception in a less constrained manner.

Textual presentations are constrained to a set of well-defined primitives (characters, symbols and mathematical operators), which are interpreted by the user in a sequential manner at a fine-grained level of detail, with each primitive examined in turn. For example, reading is a sequential low-level interpretation of the symbols on a page. The benefit of this presentation style is that it is recognised and perceived in the same way by different users and is relatively quick and easy to produce. A drawback for textual presentations is that they are not conducive to the analysis of patterns, complex data or large data sets, all of which are key characteristics of useful data mining results. Table 1 shows the results of clustering a small data set on the dimensions price and weight. Textual presentation is satisfactory in this case as the result set is small and simple, and the clustering pattern can be perceived through textual analysis. However this form of analysis is impractical for typical mining results which are both large and complex.

Table 1: Textual presentation of clustering algorithm results

CLUSTER

ITEM-ID

PRICE

WEIGHT

COLOUR

Cluster 1

102232

134

145

Blue

 

107693

98

136

Green

 

856742

87

165

Blue

Cluster 2

187456

186

145

Green

 

326478

141

154

Green

Cluster 3

788799

256

178

Red

 

275957

294

123

Blue

 

356773

321

78

Blue

 

457847

256

21

Yellow

Cluster 4

276897

218

287

Red

Graphical methods or visualisations of mining results provide more powerful forms of presentation as they are not constrained to a prespecified set of primitives as are textual presentation methods. Graphical presentations take many different forms, as the underlying data can be mapped to many different types of graphical primitives such as position, shape, colour, and size. Such diversity leads to individual visualisations being able to present many dimensions of data in a concise manner by mapping data dimensions to varied graphical primitives. By contrast, in textual presentations the data dimensions are mapped to the same textual primitive type.

Human perception and information theory (Miller, 1956) indicates that graphical presentation facilitates the search for patterns by harnessing the capabilities of the human visual system to elicit information, through visualisation, multidimensional perception, recoding, and relative judgement. Many experiments within the field of cognitive psychology have identified that regardless of sensory type (e. g., sight, taste, and smell), humans can accurately perceive differences in the stimuli to a greater extent when many parameters of that stimuli are presented. For example, in experiments by Garner, Hake and Erickson (1956), participants were presented with a series of single-dimension stimuli in the form of images each showing a point at a different position on a line. Participants were asked to label each image either from a list of possibilities or with a number from 0 to 100 indicating where to the best of their judgement the point lay on the line. Results showed that on average humans could accurately perceive approximately 10 different placements. However in experiments where the visual stimulus was increased to two dimensions (Klemmer & Frick, 1953) by the presentation of a point within a square, the level of perception rose to approximately 25 different placements. Multi-dimension perception thus suggests that graphical presentations will improve user perception due to their multi-dimensional nature. However, the relationship between dimensionality and perception has been found to be asymptotic. Above 10 or so dimensions, addition of further dimensions does not improve perception (Miller, 1956).

Recoding is the process of reorganising information into fewer chunks with more information within each chunk. This process is the means by which humans extend short-term memory (Miller, 1956). The concept of recoding suggests that it is more difficult to perceive patterns within textual presentations because of the fine-grained sequential interpretation required. This is not conducive to pattern perception as the logical units remain small, resulting in the inability to understand the underlying structure of the result set. Visual presentations present a more contiguous representation of the data that can often be interpreted as a single logical unit, providing a conducive means by which the overall structure of the data set may be examined.

Weber's law states that the "likelihood of detection [of a change] is proportional to the relative change, not the absolute change of a graphical attribute." This law indicates that a user's perception will be superior when relative judgement instead of absolute measurement is made. For example, it is easier to perceive the change in a graphical object if its original form is displayed with the newly modified representation because we can compare the difference or relative change between the two objects. Whereas it is more difficult to perceive changes when the original object is replaced by the new because no comparison is available and reliance is instead placed upon the knowledge of the object's absolute measures.

Relative judgement is a graphical capability and is a major strength of graphical presentations as it allows the users to obtain a holistic qualitative view of the result set where relative differences between items can be recognised. This qualitative view is then used to focus attention, with subsequently more focused and quantitative analysis (absolute measurement) following. This process was dubbed the visual information seeking mantra by Shneiderman (1996) and is conducive to pattern discovery as it allows the user to analyse a picture at different levels.

Although more powerful and flexible than textual presentations, graphical presentations are more difficult to create and are open to subjective interpretation, whereas textual primitives have in general a more stable interpretation. Subjective interpretation is due to the abstraction of the underlying results into graphical primitives through defined mappings. This allows the results to be presented in ways that facilitate perception of patterns and structure within the result set, but if non-intuitive mappings are used then the perception of patterns will be less predictable.



 < Day Day Up > 



Managing Data Mining Technologies in Organizations(c) Techniques and Applications
Managing Data Mining Technologies in Organizations: Techniques and Applications
ISBN: 1591400570
EAN: 2147483647
Year: 2003
Pages: 174

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net